European Wireless 2016

DRSS-based Factor Graph Geolocation Technique

for Position Detection of Unknown Radio Emitter
Muhammad Reza Kahar Aziz1,2 , Khoirul Anwar1 , and Tad Matsumoto1,3
1
School of Information Science, Japan Advanced Institute of Science and Technology (JAIST), Ishikawa, 923-1292 JAPAN
2
Electrical Engineering Department, Institut Teknologi Sumatera (ITERA), Lampung Selatan, 35365 INDONESIA
3
Centre for Wireless Communications, University of Oulu, Oulu, 90014 FINLAND
Email:{reza.kahar, anwar-k, matumoto}@jaist.ac.jp

Abstract—This paper proposes a new factor graph (FG) ge-

olocation technique utilizing differential received signal strength
(DRSS) for location detection of a single static unknown (anony-
mous) radio wave emitter. The use of DRSS-based FG (DRFG)
technique is to solve the problem of conventional received signal
strength (RSS)-based FG (RFG) technique which is unable to
estimate the position of a single static unknown radio wave
emitter without the knowledge of its transmit power knowledge.
However, in practice, the transmit power information from the
signal transmitted by the unknown target emitter is unavailable.
It should be noticed that the knowledge of the transmit power
of the target is necessary for calibration/reference of the RSS
values of training signal sent from the monitoring spots. In
this paper, we propose a new DRFG technique to eliminate the
necessity of transmit power information, and hence this technique
successfully estimates the position of an unknown radio emitter.
The performance of the proposed technique is evaluated in the
terms of root-mean-square error (RMSE). The results confirm
that the proposed technique accurately estimate the location of
unknown target, while the conventional RFG fails when the
transmit power of monitoring spots are unequal to the transmit
power of the unknown target.
I. I NTRODUCTION
Wireless geolocation has been of significant importance to
enable the location based services and applications. Hence,
this research field has attracted considerable attention in the
past two decades. The location based services and applications
have become a necessity for human life, e.g., Emergency-911
(E-911), vehicle navigation, health care, and location-sensitive
billing [1]–[3]. One of challenging problem in wireless geolo-
cation is to estimate the position of a single static unknown
(anonymous) radio wave emitter. This problem is also known
as passive radio positioning system [4], [5]. The location de-
tection of unknown radio emitter is very important for helping
people in disastrous situations, e.g., finding missing people
after disaster, the victims who are buried due to landslide,
tsunami, and/or earthquake. It is also important for monitoring
of illegal radio emitter to prevent public broadcasting from
being jammed.
In this paper we consider the use of factor graph (FG) for
detection of the position of the unknown radio target. The
FG techniques improves the accuracy of the estimate because
the probability marginalization is performed by message pass-
ing using sum-product algorithm. The FG also reduces the
DŽŶŝƚŽƌŝŶŐƐƉŽƚƐ
;‫ݔ‬௠ ǡ ‫ݕ‬௠ Ϳ
^ĞŶƐŽƌηͳ ^ĞŶƐŽƌηܰ
hŶŬŶŽǁŶdĂƌŐĞƚŵŝƚƚĞƌ
ሺܺே ǡ ܻே ሻ
ሺܺଵ ǡ ܻଵ ሻ ሺ‫ݔ‬ǡ ‫ݕ‬ሻ
^ĞŶƐŽƌηʹ
ሺܺଶ ǡ ܻଶ ሻ ƌĞĂŽĨĞƚĞĐƚŝŽŶ
tŝƌĞŽƌǁŝƌĞůĞƐƐ
tŝƌĞůĞƐƐ
&ƵƐŝŽŶĐĞŶƚĞƌ
&ĂĐƚŽƌŐƌĂƉŚƉƌŽĐĞƐƐŝŶŐŝŶĨƵƐŝŽŶĐĞŶƚĞƌ
Fig. 1. Basic structure of FG-based geolocation techniques with N = 3
sensors consisting of a single static unknown radio wave emitter, 4 monitoring
spots (monitoring spots are only required for the conventional RFG and the
proposed DRFG technique).
computational complexity of geolocation algorithm due to the
fact that the the complex global function is decomposed into
several local functions. Furthermore, the messages are in the
form of the mean and variance only, if the measurement error
follows Gaussian distribution [6], [7].
Since the FG was introduced into geolocation technique for
the first time by [8], there are many FG-based techniques1
developed according to the measurement categories, e.g.,
received signal strength (RSS) [9], [10], direction of arrival
(DOA)2 [11], time difference of arrival (TDOA) [12], and TOA
[7]. The basic structure of the FG-based technique is shown
in Fig. 1. It should be noticed that the RSS measurement has
several advantages compared to the TOA, TDOA, and DOA
measurements; which are: (i) The RSS parameter can be mea-
sured without requiring an array or direction antenna, while the
DOA measurement requires array antenna. (ii) The presence
of perfect synchronization between target and sensors, and
among the sensors is not required in the RSS measurement,
1 The term ”techniques” in this paper refer to ”geolocation technique” except
specified.
2 We use DOA acronym instead of angle of arrival (AOA) for better
expression.

ISBN 978-3-8007-4221-9 300 © VDE VERLAG GMBH, Berlin, Offenbach, Germany

 European Wireless 2016

Z^^DĞĂƐƵƌĞŵĞŶƚ ^ĞŶƐŽƌηϭ ^ĞŶƐŽƌηϮ ^ĞŶƐŽƌηϯ Hence, the proposed DRFG is able to detect the position of
ŝŶ^ĞŶƐŽƌƐ ܲ෠௪ǡଵ ܲ෠௪ǡଶ ܲ෠௪ǡଷ
WƌŽĐĞƐƐŝŶ&ƵƐŝŽŶĞŶƚĞƌ
the unknown target, while the conventional RFG fails.
Z^^&ĂĐƚŽƌEŽĚĞ II. S YSTEM M ODEL
‫ܥ‬௉
‫ܥ‬௉ ଵ ‫ܥ‬௉ ଶ ‫ܥ‬௉ ଷ

Z^^sĂƌŝĂďůĞEŽĚĞ The proposed DRFG is shown in Fig. 2. The FG is com-

ܰ௉ ோௌௌ ܰ௉ ଵ ܰ௉ ଶ ܰ௉ ଷ
Z^^&ĂĐƚŽƌ
‫ܤ‬௉ ଵǡଶ ‫ܤ‬௉ ଵǡଷ ‫ܤ‬௉ ଶǡଷ
EŽĚĞ‫ܤ‬௉
Z^^sĂƌŝĂďůĞEŽĚĞ
ܰ௉ ஽ோௌௌ
ܰ௉ ଵǡଶ ܰ௉ ଵǡଷ ܰ௉ ଶǡଷ
>ŝŶĞĂƌWůĂŶĞ&ĂĐƚŽƌ
EŽĚĞ‫ܣ‬௉ ‫ܣ‬௉ ଵǡଶ ‫ܣ‬௉ ଵǡଷ ‫ܣ‬௉ ଶǡଷ
ƐƚŝŵĂƚĞĚ'ĞŽůŽĐĂƚŝŽŶ
‫ݕ‬
ŽŽƌĚŝŶĂƚĞsĂƌŝĂďůĞEŽĚĞሺ‫ݔ‬ǡ ‫ݕ‬ሻ ‫ݔ‬ DĞƐƐĂŐĞƐĨůŽǁ
/ƚĞƌĂƚŝŽŶƉƌŽĐĞƐƐ
Fig. 2. The proposed DRFG for geolocation technique with N = 3 sensors
and P̂w,t being the input in units of watt. The new proposed nodes are marked
by yellow. The message flow arrows are shown in part to avoid over crowded
of the figure.
while the perfect synchronization is required for both the TOA
and TDOA measurements. (iii) In the RSS measurement, all
multipath can be captured by available fingerprints,3 while
the TOA, TDOA, and DOA measurements suffer from severe
multipath effects. Furthermore, In fact, the RSS measurement
has been already specified in IEEE 802.11 standard [9].
The RSS-based FG technique (RFG) in [9] uses the pattern
recognition (fingerprinting) technique. The sensors measure
the patterns or fingerprints of the RSS of training signal
sent from the monitoring spots. The position of the target
is estimated based on the resemblance between the RSS of
the signals sent from target and the training signal sent from
monitoring spots, where the position of monitoring spots are
known [9]. It is shown in Fig. 1 that the conventional RFG
technique utilizes monitoring spots.
However, the conventional RFG technique in [9] can not
estimate the position of an unknown radio emitter because the
technique requires the knowledge of the absolute value of the
transmit power. Nevertheless, in practice, the transmit power
information from the signal transmitted by the unknown target
emitter is unavailable. It should be noticed that the knowledge
of the transmit power is necessary for calibration/reference of
linear approximation process using training signal sent from
the monitoring spots.
Thus, the DRSS-based FG (DRFG) technique is proposed
in this paper to solve the problem, where the necessity for
the knowledge of the absolute transmit power of unknown
target is eliminated. The DRFG is developed by modifying
the RFG in [9]. The modification is simply by the subtraction
of RSS sample, in units of dB, between two sensors. The
DRSS samples is obtained, so that the necessity of absolute
transmit power knowledge of the unknown target is eliminated.
3 It should be noticed that, as mentioned in [9], the geolocation techniques
using RSS measurements are mainly classified into two groups: (a) pattern-
recognition (fingerprinting) techniques, and (b) model-based techniques.
posed of factor nodes denoted by square and variable nodes
denoted by circle as shown in Fig. 2. The factor node performs
processing upon the messages coming from several variable
nodes, then forwards the output messages to the destination
variable node. The FG algorithm is performed in the fusion4
center. The samples of RSS of the unknown target and of
the monitoring spots measured by each sensor are sent to the
fusion center. The DRSS and RSS samples of the unknown
target, in units of watt, are assumed to be corrupted by zero
mean Gaussian noise as in [9]. The Gaussian distribution
assumption is reasonable because of the effect of accumulation
of many independent factors [7], [9], including fading and
shadowing variations, temporal spread due to multipath, and
spatial spread.
The single static unknown (anonymous) target location
is at x = [x y]T , where (·)T is transpose function. The
knowledge of absolute transmit power of the unknown target
is unavailable. The known position of monitoring spots are at
M = [xm ym ]T . The function of the monitoring spots are to
send the training signals with knowledge of absolute transmit
power. The known sensors location are at X = [Xi Yi ]T . The
sensors measure the RSS of training signal sent from monitor-
ing spots to create the RSS profile/pattern (fingerprints). The
RSS fingerprints results are saved in database. This activity is
performed during off-line period.5 During the on-line period,
the sensors measure the RSS of signal sent from the target.
In this paper, it is assumed that the RSS samples contain
only the path-loss attenuation because the environment is in
an ideal scenario such as free space under line-of-sight (LOS)
conditions. In the case of non LOS (NLOS) conditions, this
assumption is still reasonable since the RSS samples obtained
from the signals sent from the monitoring spots and the target
are measured by group of sensors. For example, three sensors
in Fig. 1 can be represented by three groups of sensors. The
distances between sensors in a group are assumed long enough
for averaging to eliminate the shadowing variations. Hence, the
averaging results of a group of sensors are the samples with
only path-loss component.
Both of the proposed DRFG and conventional RFG tech-
niques can be used for position detection of the target6 with
either the only path-loss or shadowing fading condition. It
depends on the size of rectangular area of four monitoring
spots. The larger the size, the worse approximation obtained,
hence the linear plane created by using the RSS of signal
4 The tracking of moving target is left for future work. However, we believe
that the tracking and fusion in gelocation is connected.
5 Off-line period indicates the period before we use the proposed DRFG or
conventional RFG technique position detection of the target. On-line period
indicates the period when we perform position detection [9].
6 It is a general target. However, it should be noted that the conventional
RFG is unable for position detection of the unknown target.

ISBN 978-3-8007-4221-9 301 © VDE VERLAG GMBH, Berlin, Offenbach, Germany

 European Wireless 2016
(a) at Sensor 1. (b) at Sensor 2.
Fig. 3. RSS profile from path-loss exponent model in each sensor.
sent from the monitoring spots can approximate only path-
loss component. For example, the area size is 200 × 200 m2
or less as shown in [13]. The smaller the size, the better
approximation achieved, hence the techniques can be used in
shadowing variations under NLOS conditions.7 For example,
the area size is 6 × 6 m2 in an indoor environment as shown
in [9]. The relationship between the transmitted power of the
target and the monitoring spots is of significant importance for
this approximation.
The path-loss exponent model is used for the RSS samples
as    
4πd0 fc d the problem of conventional RFG for estimating the location
PL (d) = 20 log + 10n log , (1)
c d0
where fc is frequency carrier, d0 is reference distance, n is
path-loss exponent, and d is Euclidean distance from sensor
to target or monitoring spot [13], [14]. Figs. 3(a)–3(b) show
the RSS of path-loss exponent model with fc is 1 GHz, d0
is 100 m, n is 3 for urban area, and sensor positions at
(100, 0), (1100, 0), (600, −1000) m, however we do not show
the profile of the third sensor at (600, −1000) m in Fig. 1 to
save the space of the paper.
The RSS sample of the unknown target and monitoring
spots with path-loss attenuation is corrupted by Gaussian noise
expressed as
P̂w,i = Pw,i + nw,i , (2)
where P̂w,i is the RSS target sample in units of watt, Pw,i is
the true value of RSS target sample in units of watt, nw,i is
zero-mean Gaussian noise as measurement error of the RSS
samples in units of watt, and i, i = {1, 2, ..., N }, indicates
primary sensor index8 .
Even though the RSS-based localization is well investigated
problem, we do not use real data. Instead, in our computer
simulation, Pi , where Pw,i = 10Pi /10 , is obtained from simple
model, i.e., path-loss exponent model in (1). The proposed
DRFG technique is also not compared with other conventional
RSS-based fingerprinting techniques such as RADAR and
LANDMARC.9 These are because: (A) The conventional RFG
technique in [9] has performed simulation by using both
7 Reduction of the size of monitoring spot area to small enough size is
applied to deal with shadowing variations under NLOS conditions when
sufficient averaging can not be performed.
8 We use terms primary and secondary sensor index because DRSS param-
eter is obtained by using subtraction operation, where RSS samples value in
units of dB of the secondary sensor are subtracted from in primary sensor.
9 RADAR is a radio-frequency (RF) based system for locating and tracking
users inside buildings [15]. LANDMARC is a location sensing prototype
system that uses Radio Frequency Identification (RFID) technology [16].
ISBN 978-3-8007-4221-9 302
the shadowing model and real field measurement data in
indoor environment scenario. (B) It is shown in [9] that the
accuracy of the conventional RFG outperforms RADAR and
LANDMARC. (C) The main objective of this research is to
solve the problem of the conventional RFG technique in [9]
for detecting the position of a single static unknown radio
emitter. Hence, the comparison between the proposed DRFG
and conventional RFG is sufficient to show the effectiveness
of the proposed technique. (D) It is shown in the simulation
results that our proposed DRFG technique outperforms the
conventional RFG technique [9] when the transmit power
of the unknown target is unequal to the transmit power of
monitoring spots. Furthermore, given facts described above,
even though the simulations utilize simple model, the proposed
DRFG technique is applicable in reality.
III. P ROPOSED DRSS- BASED FACTOR G RAPH
G EOLOCATION T ECHNIQUE
The proposed technique modifies the RFG in [9] to solve
of unknown radio emitter position. The modification is simply
by performing subtraction of the RSS samples between two
sensors to obtain the DRSS samples as shown in Fig. 2.
The subtraction of RSS sample is performed in DRSS factor
node BP . Hence, the other process after obtaining the DRSS
samples follow the RFG algorithm in [9]. In this section, the
RFG in [9] is summarized as the DRFG. The analysis of the
DRSS profile, which eliminates the necessity of knowledge of
transmit power, is also presented.
The first process in the DRFG is to receive the RSS samples,
in units of watt, in RSS measurement factor node CP . After
that, the node CP converts RSS samples P̂w,i in units of watt
to P̂i dB. It should be notice that the distribution of RSS
samples, P̂i , in units of dB, shows the similarity to Gaussian
distribution. Hence, the proposed DRFG and conventional
RFG still preserve the Gaussianity assumption as shown in
[9]. The RSS samples, P̂i , in units of dB, are forwarded to
the Averaged RSS variable node NPRSS . The node NPRSS
directly forwards P̂i the node BP . The RSS samples in
units of dB from two sensors are subtracted in the node BP
resulting DRSS samples, P̂i,j , in units of dB, by performing
the operation of (3), where j, j = 2, 3, ..., N, is the secondary
sensor index. We obtain the DRSS samples derived from (2)
expressed as
P̂i,j = P̂i − P̂j . (3)
As shown in Fig. 2, there are three DRSS variables, i.e.,
P̂1,2 , P̂1,3 , and P̂2,3 . It should be noticed that P̂2,3 is a linear
combination of P̂1,2 and P̂1,3 , however, P̂2,3 is not redundant.
This is because P̂2,3 has different DRSS profile as shown in
Fig. 4(e)–4(f). Hence, P̂2,3 is useful for detection position
of the unknown target. After a set of the DRSS samples
are obtained, the node BP calculates the mean and variance
messages, in units of dB, and then forward the messages to
the DRSS variable node NPDRSS . After that, node NPDRSS
© VDE VERLAG GMBH, Berlin, Offenbach, Germany
 European Wireless 2016
(a) RSS profile at sensor 1. (b) RSS profile at sensor 2.
(d) DRSS profile at sensor 1. (e) DRSS profile at sensor 2.
Fig. 4. Linear plane and model of RSS and DRSS profiles in each sensor with ΔPT = 5 dB, the sensor positions being at (100, 0), (1100, 0), (600, −1000) m.
directly forward the messages to the linear plane LS factor
node AP . The iteration process starts from the node AP .
Before the node AP is used for the iterations, the function
at the node AP has to be set. The proposed DRFG technique
uses the error-free DRSS of four monitoring spots to obtain
the variable coefficient of the linear plane equation at the node
AP , expressed as [9]
axi,j · xm + ayi,j · ym + aPi,j · Pm,i,j = cm,i,j ,
where axi,j , ayi,j denote the coefficient of coordinate variable
x, y, respectively, xm and ym denote the position coordinate
of monitoring spots, where m = 1, 2, ..., M is the monitoring
spot index. Pm,i,j denotes the error-free DRSS of the signal
sent from monitoring spot in units of dB. The error free of
the DRSS is obtained by performing averaging over large
enough of the amount of DRSS samples. The large amount of
the samples are collected from the DRSS measurement over
long enough in time duration. aPi,j indicates the coefficient of
variable Pm,i,j . cmi,j indicates a constant with the value being
always the unity. It is assumed, for simplicity, that the four
monitoring spots are always used in the simulations so that
the target is in the middle of monitoring spots. In practice, the
target is not always in the middle of monitoring spots area.
The RSS-based Voronoi (RSS-V) technique can be used to
select the monitoring spots surrounding the target as in [10].
However, the development of the joint RSS-V and DRFG is
left for future works.
Due to the values of xm , ym , and Pm,i,j are known, hence
we can utilize the least square (LS) algorithm to obtain the
coefficients of linear equations at the node AP . The detailed
explanation of the use of LS for obtaining the coefficients of
the variables axi,j , ayi,j , and aPi,j , can be found in [9]. The
ISBN 978-3-8007-4221-9
(c) RSS profile at sensor 3.
(f) DRSS profile at sensor 3.
coefficients complete the final linear equations as
axi,j · x + ayi,j · y + aPi,j · Pi,j = ci,j , (5)
where x and y denotes the unknown target position, Pi,j is
the DRSS of unknown target in unit of dB, and ci,j is the
constant with value being unity. The detail derivation formula
of (5) in terms of mean and variance can be found in [9].
In this paper, we summarize the equations/operations used
in this algorithm in Table. I, with the most left column being
(4)
the message flow between the nodes, h being general sensor
index, m being the mean, and σ 2 being the variance. For
example, mx and my indicates the mean values from the
nodes x and y, and mP̂ indicates the mean of the samples
from measurement. The final position estimate of the unknown
target emitter is taken from the mean value, mΛ .
It is also shown in Table. I that the proposed technique
requires simple arithmetic operations. As shown in [9], the
computational complexity of the conventional RFG is linearly
proportional to N . Hence, the computational complexity of the
proposed technique is also linearly proportional to N , because
the DRFG only introduces one additional subtraction operation
to the conventional RFG.
We compare both the RSS the DRSS profiles at 3 sensors
calculated using (1) with transmit power gap between the
target and monitoring spots, ΔPT , being 5 dB. Both the
approximated DRSS and RSS profiles are calculated by using
(5). The simulation setup is detailed later in the Section III.
As shown in Figs. 4(a) – 4(f), there are gaps between the RSS
profile of path-loss mode and the approximated linear plane of
RSS profile at all sensors, because there is the gap of transmit
power between the unknown target and the monitoring spots.
Hence, the conventional RFG fails in estimating the position
of unknown radio emitter. On the contrary, the path-loss
plane of DRSS has intersection with the approximated DRSS
303 © VDE VERLAG GMBH, Berlin, Offenbach, Germany
 European Wireless 2016

0 500
0
-200 -200
0
Y (meter)

Y (meter)

Y (meter)
-400 -400
-500
-600
-600
-800 Sensor
-1000 Sensor
RSS-based FG RSS-based FG
-800 DRSS-based FG DRSS-based FG
Unknown target
-1000 Sensor Monitoring spot Monitoring spot
Target Target
0 500 1000 -1000 -1500
0 500 1000 1500 0 500 1000 1500 2000
X (meter) X (meter) X (meter)
(a) 1, 000 of 10, 000 targets and 3 sensors. (b) Target at (650, −500) m. (c) Target at (330, −550) m.

Fig. 5. Accuracy of the unknown target detection for many locations confirmed by the trajectory analyses with ΔPT = 5 dB.

TABLE I. T HE O PERATIONS R EQUIRED FOR E ACH N ODE IN T HE P ROPOSED DRFG.

Message Flow Samples (ˆ) and/or (Means, Variances)
(Nodes) Inputs Outputs Remarks
C Ph → N Ph P̂w,h , h = {i, j} P̂h , h = {i, j} Similar to [7], [9]
NPh → BPi,j P̂h , h = {i, j} P̂h , h = {i,  j}  New proposed
BPi,j → NPi,j P̂i and P̂j P̂i,j = P̂i − P̂j and mP̂i,j , σP̂2 New proposed
    i,j

NPi,j → APi,j m , σ2 mP̂i,j , σP̂2 Similar to [7], [9]

 P̂i,j P̂i,j  i,j

m , σ2 ((ci,j − ayi,j myi,j − aPi,j mP̂i,j )/axi,j ,

APi,j → x  P̂i,j P̂i,j Similar to [9]; Iteration process is only
my , σ 2 (a2yi,j σy2i,j + a2Pi,j σP̂2 )/a2xi,j ))
performed between nodes APi,j and x,
 i,j yi,j 
i,j

m , σ2 ((ci,j − axi,j mxi,j − aPi,j mP̂i,j )/ayi,j , y. Initial values are set for mxi,j , myi,j ,
APi,j → y  P̂i,j P̂i,j σx2i,j , σy2i,j
mxi,j , σx2i,j (a2xi,j σx2i,j + a2Pi,j σP̂2 )/a2yi,j ))
i,j
   
x → APi,j mk , σk2  Similar to [7], [9]; Iteration process;
σk2 l=k m l
, σ 2
=  1
y → APi,j k = l σl2 k 1
l=k σ 2 For simplicity, let replace (·)i,j to (·)k
 l 
   mk 2 1 Similar to [7], [9]; Iteration converges;
x and y mk , σk2 2
mΛ = σΛ k σ2 , σΛ = 
k
1
k σ2
k
For simplicity, let replace (·)i,j to (·)k

profile around the unknown target, because the necessity of the
knowledge of transmit power is eliminated by the subtraction
in the node BP . Hence, the DRFG successfully estimates the
position of unknown target radio emitter.
IV. S IMULATION R ESULTS
To verify the performance of the proposed technique, we
conducted a series of computer simulations. One target is
randomly chosen from area width of 600 × 600 m2 in each
10, 000 target positions performed in the computer simula-
tions. 100 samples of RSS and DRSS are processed in 30
times of iteration for each trial. The sensors position are
at (100, 0), (1100, 0), (600, −1000) m in (X, Y ) coordinate,
where the sensors area width is 1, 000 × 1, 000 m2 as shown
in Fig. 5(a). The root-mean-square error (RMSE) performance
is evaluated by the signal-to-noise power ratio (SNR) from 0 to
45 dB. The gap of transmit power between the unknown target
radio emitter and the monitoring spots is 5 dB. For simplicity,
the RSS and DRSS measurements are assumed to be corrupted
by the measurement error having the same variance in each
sensor. The path-loss exponent model in (1) is used to create
the RSS profile, where the set-up of the exponent path-loss
variables are set in Section II. The monitoring spots area set
in this simulation is 200 × 200 m2 according to [13].
Figs. 5(b) – 5(c) show how the proposed DRFG is very
accurate in detection, even though the location of the unknown
target is changed many times. The accuracy is confirmed via
the trajectory analysis shown in the figures. The initial point
can be set from arbitrary value, however we set the initial
value of nodes x and y at (0, 0) m. It is shown in Figs. 5(b)
– 5(c) that the conventional RFG fails to reach the unknown
target radio emitter. This is because as we discuss in Section
III that the RSS profile of exponent path-loss model has a
gap ΔPT = 5 dB to the approximated RSS profile created by
linear plane of (5). On the other hand, the trajectories of the
DRFG shows that the proposed technique successfully reach
the unknown target.
The Figs. 6 and 7 show that the proposed technique for any
ΔPT values is not only able to detect the location of unknown
target radio emitter, but it also provides very accurate detection
after it converges around 50 iterations with RMSE of around
4.5 and 3.7 m, at SNR of 10 dB and above, respectively. It
is also shown that the high accuracies with RMSE of around
5 and 4 m even have been achieved in 30 and 40 iterations,
respectively, at SNR of 15 dB. Hence, the proposed DRFG
technique achieves high and stable accuracy for any transmit
power gap values between the target and monitoring spots, but
it requires more iteration.

ISBN 978-3-8007-4221-9 304 © VDE VERLAG GMBH, Berlin, Offenbach, Germany

 European Wireless 2016

3
10 paper that when the transmit power of target and monitoring
spots are unequal, the proposed technique successfully esti-
mates the unknown radio emitter with high accuracy around
RFG ΔPT = 5 dB
RMSE of 3.7 m for average distance between the targets and
RFG ΔP = 3 dB
10
2 T sensors being around 664 m, while the conventional RFG fails
RFG ΔPT = 1 dB
RMSE (meter)

RFG ΔP = 0 dB
to detect the location of the unknown target. The transmit
T
DRFG ΔP = 5 dB
power information of the unknown target is no longer required
T
DRFG ΔP = 3 dB because it can be replaced by performing subtraction of one
T

10
1 DRFG ΔP = 1 dB
T
sensor’s RSS samples from another sensor’s RSS samples.
DRFG ΔP = 0 dB
T
This technique provides high accuracy and low computational
complexity detection, which is suitable for future geolocation
technique.
10
0 VI. ACKNOWLEDGEMENT
0 20 40 60 80 100
Iteration number This research is in part supported by Koden Electronics Co.,
Fig. 6. RMSE vs. iteration number with 10, 000 target positions, SNR of 15 Ltd.
dB, and 100 samples.
10
3 R EFERENCES
RFG ΔP = 0 dB
T
[1] J. James J. Caffery and G. L. Stuber, “Overview of radiolocation in
RFG ΔP = 1 dB
T CDMA cellular systems,” IEEE Communications Magazine, vol. 36,
RFG ΔP = 3 dB
T no. 4, pp. 38–45, April 1998.
RFG ΔP = 5 dB [2] K. Pahlavan, X. Li, and J.-P. Makela, “Indoor geolocation science and
T
2 DRFG ΔP = 0 dB technology,” IEEE Communication Magazine, vol. 40, pp. 112–118,
10 T
RMSE (meter)

DRFG ΔPT= 1 dB February 2002.

DRFG ΔPT= 3 dB
DRFG ΔPT= 5 dB
1
10
0
10
0 10 20 30 40
SNR (dB)
Fig. 7. RMSE vs. SNR with 10, 000 target positions, 100 samples, and 30
iterations.
On the other hand, Figs. 6 and 7 show that when the target
and monitoring spots have equal transmit power, the accuracy
of conventional RFG technique, with RMSE of 1.9 and 1.6
m, at SNR of 15 dB and above, respectively, outperforms the
proposed DRFG technique. This is because the linear plane
approximation of RSS parameter has better shape than the
DRSS parameter as shown in Figs. 4(a) – 4(f). However, when
there are transmit power gaps, 1, 3, and 5 dB, the accuracy of
the conventional RFG technique drop significantly to 57, 171,
and 286 m, respectively, at SNR of 15 dB. The performance
trade-off shows that the accuracy of the proposed DRFG
outperforms the conventional RFG for position detection of
the unknown target, while the conventional RFG has better
accuracy with having the absolute transmit power knowledge
of the target. Opposite to the DRFG curve, the RFG curve
shows the best accuracy in SNR of 0 dB because the gap 5 dB
of transmit power is very big. Hence, the high noise power in
0 dB helps the RFG for better accuracy than other SNR values
even tough the overall accuracy is still low.
V. C ONCLUSION
The new technique of DRSS-based FG (DRFG) geolocation
algorithm has been presented in this paper. It is shown in this
[3] Y. Zhao, “Standardization of mobile phone positioning for 3G systems,”
IEEE Communication Magazine, vol. 40, pp. 108–116, July 2002.
[4] Y. Wang and K. Ho, “TDOA source localization in the presence of syn-
chronization clock bias and sensor position errors,” Signal Processing,
IEEE Transactions on, vol. 61, no. 18, pp. 4532–4544, Sept 2013.
[5] Q. Liang, B. Zhang, C. Zhao, and Y. Pi, “TDoA for passive localization:
Underwater versus terrestrial environment,” Parallel and Distributed
Systems, IEEE Transactions on, vol. 24, no. 10, pp. 2100–2108, Oct
2013.
[6] F. R. Kschischang, B. J. Frey, and H.-A. Loeliger, “Factor graphs and the
sum-product algorithm,” IEEE Trans. on Information Theory, vol. 47,
50
no. 2, pp. 498–519, February 2001.
[7] J.-C. Chen, Y.-C. Wang, C.-S. Maa, and J.-T. Chen, “Network side
mobile position location using factor graphs,” IEEE Trans. on Wireless
Comm., vol. 5, no. 10, pp. 2696–2704, October 2006.
[8] J.-C. Chen, C.-S. Maa, and J.-T. Chen, “Factor graphs for mobile
position location,” in Proc. IEEE International Conference on Acoustics,
Speech, and Signal Processing (ICASSP) 2003, vol. 2, April 2003, pp.
393–396.
[9] C.-T. Huang, C.-H. Wu, Y.-N. Lee, and J.-T. Chen, “A novel indoor RSS-
based position location algorithm using factor graphs,” IEEE Trans. on
Wireless Comm., vol. 8, no. 6, pp. 3050–3058, June 2009.
[10] M. R. K. Aziz, Y. Lim, and T. Matsumoto, “A new wireless geolocation
technique using joint RSS-based Voronoi and factor graph,” 9th Asia
Modelling Symposium (AMS) 2015, pp. 132–136, Sep 2015.
[11] B. Omidali and S. A.-A. B. Shirazi, “Performance improvement of
aoa positioning using a two-step plan based on factor graphs and the
gauss-newton method,” in 14th International CSI Computer Conference
(CSICC 2009), September 2009, pp. 305–309.
[12] C. Mensing and S. Plass, “Positioning based on factor graphs,” EURASIP
Journal on Advances in Signal Processing, vol. 2007, no. ID 41348, pp.
1–11, April 2007.
[13] M. R. K. Aziz, K. Anwar, and T. Matsumoto, “Monitoring spot con-
figuration of RSS-based factor graph geolocation technique in outdoor
WSN environment,” in IEICE General Conference 2015, March 2015.
[14] Y. S. Cho, J. Kim, W. Y. Yang, and C.-G. Kang, MIMO-OFDM Wireless
Communications with MATLAB. Wiley, 2010.
[15] P. Bahl and V. Padmanabhan, “Radar: an in-building rf-based user
location and tracking system,” in INFOCOM 2000. Nineteenth Annual
Joint Conference of the IEEE Computer and Communications Societies.
Proceedings. IEEE, vol. 2, 2000, pp. 775–784 vol.2.
[16] L. Ni, Y. Liu, Y. C. Lau, and A. Patil, “Landmarc: indoor location
sensing using active rfid,” in Pervasive Computing and Communications,
2003. (PerCom 2003). Proceedings of the First IEEE International
Conference on, 2003, pp. 407–415.

ISBN 978-3-8007-4221-9 305 © VDE VERLAG GMBH, Berlin, Offenbach, Germany